A centralized photovoltaic power station whole life cycle intelligent management system

Abstract

The application relates to the technical field of photovoltaic power station management, and particularly discloses a centralized photovoltaic power station whole life cycle intelligent management system. The system comprises a model management module, a data service module, a decision guidance module and a state tracing module. The digital twin body constructed by the model management module can dynamically reconstruct the model granularity and depth according to the life cycle stage of the power station; the data service module standardizes and encapsulates the whole cycle heterogeneous data according to a predefined data structure template; the decision guidance module generates strategy instructions based on the simulated state of the digital twin body; and the state tracing module records all state changes as a linear event sequence for tracing and auditing. Through the cooperation of the modules, the application realizes the whole process, self-adaptive and traceable integrated management of the photovoltaic power station from planning to decommissioning.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic power plant management technology, and more specifically, to a centralized intelligent management system for the entire life cycle of a photovoltaic power plant. Background Technology

[0002] With the transformation of the global energy structure, the construction scale of centralized photovoltaic power plants is increasing daily. Their full life-cycle management, encompassing the entire process from planning, construction, operation, maintenance to decommissioning, is crucial for ensuring the long-term economic benefits and operational safety of the power plants. Achieving efficient coordination and precise decision-making at each stage has become key to enhancing the overall asset value of the power plants.

[0003] Currently, most existing management systems adopt an architecture of phased independent construction or simple stacking of functional modules. There are widespread problems of inconsistent data standards and incompatible models among simulation software in the planning phase, project management tools during construction, data acquisition and monitoring control systems during operation, and diagnostic platforms during maintenance, resulting in severe data silos. This leads to fragmented information across the entire lifecycle and a lack of coherent decision-making. For example, simulation models from the planning phase cannot be directly used to guide refined operation and maintenance, while the massive amounts of data accumulated during the operation phase are difficult to use to optimize the initial design. More significantly, the digital models of existing systems are mostly static or pre-set, lacking adaptive evolution capabilities. They cannot dynamically adjust their structure and granularity according to changes in the physical state of the power plant and the business phase, resulting in a significant reduction in the applicability and guiding value of the models in the latter half of the lifecycle, and also posing enormous difficulties for accident tracing and root cause analysis.

[0004] Therefore, a centralized intelligent management system for the entire life cycle of photovoltaic power plants is proposed to address the above problems. The core issue to be solved is: how to break down the barriers between different management stages and build a unified management framework that can evolve adaptively and ensure the consistency and traceability of data throughout the entire process, so as to support cross-stage collaborative optimization and accurate decision-making. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, embodiments of the present invention provide a centralized photovoltaic power plant full life cycle intelligent management system to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: Preferably, a centralized photovoltaic power station full life cycle intelligent management system includes: A model management module is used to build and maintain a digital twin that is mapped to a physical photovoltaic power plant in real time. The model management module is configured to perform a model reconstruction operation in response to an instruction representing a life cycle phase transition. The reconstruction operation dynamically changes the set of simulated objects and the level of detail of their attributes in the digital twin. The life cycle phases include planning, construction, operation, maintenance and decommissioning phases. A data service module is used to standardize and encapsulate heterogeneous data from various life cycle stages according to a predefined data structure template corresponding to the photovoltaic power station field. The data structure template requires each data entity to be bound to a semantic type label taken from a closed type dictionary and a timestamp accurate to the second, and provides the encapsulated data stream to the model management module. A decision guidance module is used to generate strategy instructions to guide work at specific lifecycle stages based on the simulated state presented by the digital twin after inputting the encapsulated data stream. A state tracing module is used to record all external instructions, data update events and internal policy instructions received by the system in their order of occurrence as an unmodifiable linear event sequence, and to reproduce the complete state of the digital twin at any specified historical moment by replaying the event sequence sequentially from the initial state.

[0007] Preferably, the instructions used by the model management module to perform the model reconstruction operation are automatically triggered by any of the following conditions: the preset milestone event marker in the business process data is identified; or the decision guidance module issues a phase transition suggestion and obtains confirmation through the authorized interface; or the data service module reports a structural change in the component topology of the physical power station involving a preset percentage or more of the number of nodes.

[0008] Preferably, the model reconstruction operation specifically manifests as follows: S11. When the phase changes from planning to construction, in the digital twin, the set of simulated objects is expanded from the site-level geographic space and electrical boundaries to include the location coordinates and electrical connection relationships of key equipment. S12. When the phase transitions from construction to operation, the digital twin further refines the attributes of the key equipment from static parameters to include real-time operating parameters and health status indicators calculated based on performance data.

[0009] Preferably, when generating policy instructions, the decision guidance module calls the historical event sequence provided by the state tracing module and optimizes policy generation by performing the following steps: S21. Extract a set of key feature vectors from the current state of the digital twin; S22. Retrieve historical state points and their subsequent events with similar feature vectors in the historical event sequence; S23. Analyze the strategies executed at the historical state points and the impact of the subsequent event sequences on the long-term performance indicators of the system. S24. Based on the analysis results, adjust the generation logic of the current strategy instructions to avoid the negative effects that have been shown in the historical strategies or to enhance their positive effects.

[0010] Preferably, the predefined data structure template in the data service module has a closed type dictionary that covers all core entity types from photovoltaic modules and inverters to construction logs and maintenance work orders, and ensures consistent understanding and seamless integration of cross-stage data at the semantic level through the semantic type tags.

[0011] Preferably, the decision guidance module is configured to perform the following simulation optimization steps before generating the final strategy instruction: S31. Based on the current state of the digital twin, generate at least two candidate strategies; S32. Using each candidate strategy as an input event, drive the digital twin to perform virtual execution, and obtain a set of system status indicators after virtual execution. The indicators include at least system efficiency, security risk level and expected benefits. S33. Based on the preset priority rules, compare the system status indicators of each group to select the final output strategy instruction.

[0012] Preferably, the preset priority rule is configured as follows: S41. First, assess the security risk level of all candidate strategies and select a set of strategies with a risk level lower than a preset threshold. S42. In the filtered set of strategies, the candidate strategy that maximizes the improvement in system efficiency is selected as the final output.

[0013] Preferably, when the model management module detects that the power plant has entered the decommissioning stage, the model reconstruction operation performed includes: S51, Remove the simulation of real-time operating parameters of the equipment; S52. Associate the equipment object with the asset residual value assessment model and material recycling and processing path information; S53. Shift the focus of the model's calculations from real-time power generation performance to the simulation and evaluation of the value of the inventory and the subsequent utilization plan of the site.

[0014] Preferably, the event sequence recorded by the state tracing module constitutes the unique source of facts for the system state. Based on this, the audit query function can reverse locate and output all upstream related events that led to the result and their precise timestamps for any policy instruction or data change, forming a complete causal chain.

[0015] The technical effects and advantages of this invention are as follows: Compared to existing technologies, this invention constructs a digital twin that can be dynamically reconstructed in response to lifecycle phase transitions and sets specific model granularity switching rules. Based on predefined milestone events or structural change instructions, the system automatically adjusts the set of simulated objects and attribute depth within the digital twin, for example, from the site layout evolution in the planning phase to equipment-level parameters in the operation phase. This approach ensures that the virtual model always closely matches the actual operational status of the power plant, providing appropriate simulation support for different stages. This overcomes the mismatch problem of traditional static models throughout the entire lifecycle, significantly improving the accuracy and timeliness of decision analysis.

[0016] Compared to existing technologies, this invention introduces a state tracing mechanism that uses event sequences as the sole source of facts, recording all operations that lead to system state changes as an immutable linear event log. When generating strategies, the decision-making guidance module can invoke this complete event sequence, using feature matching to trace back strategies and their long-term impacts under similar historical states, thereby optimizing the current decision. This process transforms discrete historical data into continuous, causally clear learning samples, enabling the system to learn from past experience, effectively avoid known risks, enhance the scientific rigor and foresight of the decision-making process, and overcome the limitations of relying on instantaneous data for judgment.

[0017] Compared to existing technologies, this invention employs a standardized data template based on a closed-type dictionary to uniformly encapsulate multi-source heterogeneous data throughout its entire lifecycle. This template mandates that each data entity be assigned precise semantic tags and timestamps, fundamentally ensuring semantic consistency across stages and protocols. The data service module thereby provides semantically consistent data streams to the upper-layer model, eliminating analytical errors caused by data ambiguity. This lays a solid foundation for achieving seamless data fusion and interaction throughout the entire process from planning and design to decommissioning and recycling, solving the long-standing problem of data silos. Attached Figure Description

[0018] Figure 1 This describes the overall workflow of the method of the present invention.

[0019] Figure 2 This is the core processing and data analysis workflow of the present invention.

[0020] Figure 3This is the risk assessment and decision-making process of the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Example

[0022] As attached Figures 1 to 3 The centralized photovoltaic power plant full life cycle intelligent management system shown has a digital twin maintained by its model management module, which is a hierarchical model composed of a parametric component library.

[0023] After the data service module receives multi-source data through the adapter, it calls its built-in semantic parser and standardized templates to process the data. It maps the data into a unified JSON-LD format, based on a predefined, closed photovoltaic domain ontology.

[0024] The decision guidance module consumes this standardized data through a rules engine and simulation interface. The state tracing module encapsulates any operation that causes a state change as an event object and appends it strictly in chronological order to an append-only event store, thus forming an immutable, linear historical record that allows for the reconstruction of any historical moment by replaying the event sequence.

[0025] Furthermore, the model refactoring is triggered by a continuously running condition monitoring engine that listens to three asynchronous input streams in parallel.

[0026] For business process data, it uses keyword matching and a state machine mechanism for judgment. For suggestions from the decision guidance module, the engine receives them through an authorization API interface, and only after receiving a confirmation signal will the suggestion be transformed into an executable trigger instruction.

[0027] For physical topology changes, the data service module periodically performs a difference analysis on the equipment connection diagram of the power plant, using graph theory algorithms to calculate the node differences between the current topology and the previous version. Specifically, it calculates the Jaccard difference coefficient between the two graph versions, with the formula as follows: In this formula, A set of unique identifiers (such as IDs) for all device nodes (including photovoltaic strings, inverters, combiner boxes, etc.) in the previous version of the topology; A set of unique identifiers (such as IDs) for all device nodes (including photovoltaic strings, inverters, combiner boxes, etc.) in the previous version of the topology. Represents the set of unique identifiers for all device nodes in the current version of the topology; This represents the number of common device nodes that exist in both versions; This represents the total number of unique device nodes in both versions. This coefficient... It visually reflects the proportion of change in the set of device nodes. Once The value exceeds a preset threshold (e.g.) Upon receiving this information, a structural change alert will be generated immediately. Furthermore, the model reconstruction operation is specifically manifested as a structured editing process of the digital twin model tree. When an instruction to transition from planning to construction is triggered, the reconstruction engine performs model refinement operations, instantiating key equipment nodes in the model tree based on the imported equipment list and design drawings, and maintaining their electrical connections through an adjacency list data structure.

[0028] When the command to transition from construction to operation is triggered, the engine executes a second round of in-depth operations, traversing all device nodes and dynamically loading and attaching runtime behavior scripts. For example, a microservice is attached to the inverter node, continuously receiving its operating current and temperature data. It first uses a weighted moving average (WMA) method to smooth the raw data, where the weight allocation follows the time decay principle, meaning that data points closer to the current time have higher weights. The specific weights are calculated using the formula... Calculation, where The sequence number of the data point within the time window (current time). ), This refers to the size of the time window, the value of which can be configured based on the parameter characteristics and monitoring requirements (for example, for slowly varying parameters such as temperature). It should take 1 hour; for transient parameters such as current, (Corresponding to 1 minute). The system maintains a fixed length of... A first-in, first-out (FIFO) queue is used to store the data within the window. Each time a new data point is added, the oldest data point is removed, and the weighted moving average is recalculated in real time. Subsequently, the smoothed data is compared with the rated operating condition threshold range preset according to the equipment model to calculate its deviation.

[0029] Ultimately, the 'health status index' was quantified into one... Fractions of -100 are obtained using the formula The calculation yielded the following result. For the first The parameter values ​​(such as current) obtained from the next sampling. This is the rated threshold for this parameter. The duration for which the sampled value continuously exceeds the limit. Total observation time This is a standardized coefficient preset based on the equipment type, used to normalize the deduction values ​​to... -100 range.

[0030] Furthermore, the decision guidance module introduces an optimization layer based on case-based reasoning (CBR) when generating strategies.

[0031] Its workflow begins with feature extraction, which involves filtering key indicators from high-dimensional state data to form a numerical feature vector. The feature vector is specifically composed of .

[0032] in, The deviation rate of total power generation relative to the predicted value; This is the standard deviation of the current values ​​of all strings at the current moment, used to quantify the degree of current imbalance; This represents the temperature difference between the ambient temperature and the standard test conditions. The highest component temperature at all monitoring points; This represents the total number of currently active alerts.

[0033] Subsequently, the system uses Position-Sensitive Hash (LSH), a typical ANN algorithm, to find similar states in the historical event sequence. Specifically, it maps high-dimensional feature vectors to a set of hash tables and quickly retrieves potential nearest neighbors by comparing the codes of hash buckets, thereby reducing the search time complexity from [previous level] to [new level] with a small loss of precision. The performance is significantly reduced to sublinear levels. Next, the event chains within an analysis period are mined to quantitatively assess the long-term consequences of the strategies adopted under each similar state at that time. Finally, based on this analysis, the decision logic uses a performance feedback-based weight update mechanism to optimize its internal strategy selection probabilities.

[0034] Specifically, for each historical case that was recalled Its corresponding historical strategy It will be assigned a weight The weight is calculated as follows: .in, This is a case study. Similarity to the current state; It is a quantitative indicator of negative effects, such as the case study. The percentage of power generation loss caused by the adopted strategy over a subsequent period; It is greater than The penalty factor.

[0035] This factor is adaptive through an offline feedback loop: the system periodically (e.g., monthly) collects a data set containing... A dataset of historical decisions ,in Let be the predicted negative effect at the time of decision-making, and be the actual negative effect observed after the decision. The system minimizes the loss function:

[0036] Update Its gradient descent update formula is: in Let the learning rate be the gradient formula:

[0037] Iterate until convergence or the maximum number of iterations is reached.

[0038] The core of the data service module, which includes the percentage deviation of total power generation from the predicted value, regional current imbalance, ambient temperature and rated temperature difference, highest component temperature, and number of alarms, lies in its ontology-based data standardization capabilities.

[0039] When raw data (e.g., a data frame read via the Modbus TCP protocol) is received, the paradigm engine first looks up the data in a device location mapping table based on its source address (e.g., IP and port) and register start address. This table is created and validated during the power plant construction phase, and each record explicitly defines: device IP, register address, corresponding device ID, parameter type (associated with a closed type dictionary), and data parsing rules defining the conversion method from raw data to engineering values. For a 16-bit signed integer register value... Its engineering value ,in This is a scaling factor (e.g., 0.1). Offset (e.g.) Based on this, the engine completes the conversion from raw binary data to key-value pairs with clear semantics, and outputs standardized data objects with a unified structure and clear semantics according to the JSONSchema template corresponding to the data type.

[0040] Furthermore, the simulation-based optimization process of the decision-guiding module is a typical simulation-based optimization loop.

[0041] The process first generates a diverse space of candidate policies. Then, the simulation scheduler injects each policy as a set of initial conditions and rules into the digital twin's simulation environment, driving the model to rapidly simulate based on physical formulas (such as photovoltaic power generation models) and calculate a series of outcome metrics. The policy evaluator then intervenes, first performing a safety filter: A system based on production rules is applied for security filtering. Each rule takes the following form: IF<condition>THEN<action>.

[0042] For example, a rule can be defined as: IF In the simulation results, any inverter radiator temperature > 85°C THEN Strategy risk level = HIGH; Strategy feasibility = FALSE. After the simulation ends, the system iterates through all predefined safety rules, matches and evaluates the outputs of all candidate strategies, and any strategy that makes the rule conditions true will be immediately marked as high risk and removed from the candidate set.

[0043] In the pool of policies that have passed the safety screening, the evaluator sorts them according to the principle of multi-objective optimization. Pareto optimality is often used to find a set of non-dominated solutions, or the final policy is selected according to a preset single objective (such as maximizing system efficiency).

[0044] Furthermore, the model reconstruction triggered during the decommissioning phase is a paradigm shift. The system first performs resource cleanup, severing connections with all real-time data sources.

[0045] Next, the asset mapping process is initiated, associating each physical device node in the model with its specific attributes during the decommissioning phase. The attribute values ​​are automatically populated by querying the built-in residual value assessment model.

[0046] This model is a portfolio valuation model, and its formula is: In this formula, Represents the estimated recovery value; Represents the original value of the equipment; This represents the annual depreciation rate determined by the type and quality of the equipment. Indicates the service life; It is a technical status adjustment coefficient, which is directly mapped from the device's final health status index (HealthScore, H) through a piecewise linear function: when hour, ;when hour, ;when hour, ; The value of representative materials is calculated as follows: That is, recyclable materials in the equipment A type of material, its weight Multiply by the current market recycling price per unit Then sum them up. Where the weight... Instead of on-site weighing, the weight is obtained by querying a pre-built 'equipment and materials list library'. This library is built based on the product specifications when the equipment is received into the library. Its core table structure includes fields such as: equipment model, material type (e.g., 'silicon', 'glass', 'aluminum'), and theoretical weight. (Unit: kg). When performing residual value assessment, the system queries this database by equipment model and automatically obtains the weight data of all relevant materials for summation calculation.

[0047] Furthermore, the event sourcing-based auditing function can reconstruct the complete causal chain of any decision. The auditing engine strictly follows the event sourcing design paradigm, tracing back to the direct cause of an event. This process is achieved by mandating the reading of the mandatory causality_event_id (causal event ID) field from the event payload. Every state change event must contain this field, which either points to a unique ID of another event (indicating its causal source) or is NULL (indicating that it is a system-initial event or directly triggered by external input).

[0048] For the latter, the event payload must fully record the trigger source (such as user ID, sensor ID) and a snapshot of the raw data. The causal chain is constructed by recursively tracking non-empty causality_event_id pointers.

[0049] The process begins with a target event and uses a depth-first search algorithm to traverse the event storage graph in reverse order along the causality_event_id pointer until a root event with causality_event_id set to NULL is encountered. All events along the traversal path are ordered in ascending order of their timestamps, thus forming a complete causal chain.

[0050] The audit report will present this causal chain along with the complete payload data for each event in the chain, providing irrefutable, machine-readable decision-making support.

[0051] The complete workflow of this invention is as follows: Step 1: System Initialization and Data Access After system startup, the data service module loads a predefined device location mapping table and a photovoltaic domain ontology dictionary. Subsequently, it continuously receives multi-source heterogeneous data through various adapters (such as ModbusTCP and HTTPAPI). When a piece of raw data (e.g., a data frame from an inverter with IP address 192.168.1.10 and a register starting at address 0x0000) arrives, the data service module queries the mapping table, parses it as the DC-side current of inverter A, and converts it into a standardized data object according to a JSONSchema template, for example: {entity_id:INV_A, property:dc_current, value:12.5, unit:A, timestamp:1621234567}. This object is published to the system's internal message bus, and simultaneously, the state tracking module immediately records this data update event to the immutable event store.

[0052] Step 2: Condition Monitoring and Refactoring Triggering The condition monitoring engine in the model management module continuously listens to the message bus. It receives and analyzes standardized data streams, business process events (such as work order status updates), and suggestions from the decision guidance module. The engine makes judgments based on preset rules: if a project completion marker is detected in the business process data, a construction → operation phase transition instruction is generated; if the decision guidance module suggests entering de-load operation and this is confirmed, a normal → de-load mode transition instruction is generated; if the data service module calculates the Jaccard difference coefficient of the topology… ,Discover If either condition is met, a structural change instruction is generated. Once either condition is met, the engine issues a reconstruction instruction with a specific target state to the model reconstruction engine.

[0053] Step 3: Dynamic Reconstruction of the Digital Twin After receiving an instruction, the model refactoring engine loads the corresponding target configuration file from the model configuration library. For example, for the "Build → Operation" instruction, the configuration requires deepening the granularity of all inverter models from location and connectivity to real-time operation and health status. The refactoring engine traverses the digital twin, dynamically attaching a health status calculation microservice to each inverter node. Once activated, this service immediately begins receiving standardized data streams and applies a weighted moving average method. The input current is smoothed, and then the smoothed data is compared with the rated threshold, according to the formula: Ultimately, the 'health status index' was quantified into one... Fractions of -100 are obtained using the formula The calculation yielded the following result. For the first The parameter values ​​(such as current) obtained from the next sampling. This is the rated threshold for this parameter. The duration for which the sampled value continuously exceeds the limit. Total observation time This is a standardized coefficient preset based on the equipment type, used to normalize the deduction values ​​to... -100 range; The refactoring completion event is recorded by the state tracing module.

[0054] Step 4: Strategy Generation and Simulation Optimization The decision-guiding module initiates a policy generation cycle based on the reconstructed digital twin with refined states. It first extracts feature vectors from the current state. Subsequently, the case reasoning optimization layer is invoked, using an approximate nearest neighbor search algorithm to find the most similar events in the historical event sequence. Each historical state case. For each case... The system calculates its strategy weights. Among them, the penalty factor These values ​​are obtained through offline gradient descent optimization. Based on these weights, the module generates... - The system identifies several candidate strategies and drives a digital twin to simulate the execution of each strategy over the next 24 hours, outputting metrics such as system efficiency and security risks.

[0055] Step 5: Strategy Evaluation and Decision Implementation After receiving the simulation results, the policy evaluator first performs security filtering, eliminating all policies that would cause any security risk level to exceed the lower limit. Then, from the remaining candidate policies, it selects the final policy according to preset priority rules (e.g., prioritizing policies that offer the greatest improvement in system efficiency). The decision guidance module will (For example, cleaning array B05 from 14:00 to 16:00 today) is issued as a policy instruction to the message bus. The status tracing module records the policy decision and instruction issuance events. This instruction is issued to the corresponding subsystem or device for execution through the executor interface.

[0056] Step 6: Continuous Tracking and Closed-Loop Optimization After the physical power plant executes the command, the resulting state changes (such as increased power generation after cleaning) are captured as new data by the data service module, restarting the process from step one. All events generated throughout the process constitute a complete causal chain. The system periodically (e.g., weekly) initiates an offline learning process: It aggregates the actual effects of all executed strategies and uses gradient descent to adjust the penalty factor in case reasoning. This minimizes the error between predictions and actual negative effects, thereby enabling the self-evolution of decision-making logic. When the power plant enters the decommissioning phase, the model management module will automatically trigger a final reconstruction, switching the model to decommissioning assessment mode and utilizing formulas... Automatically generate power plant decommissioning assessment reports.

[0057] Finally, the following points should be noted: First, in the description of this application, it should be noted that, unless otherwise specified and limited, the terms installation, connection, and link should be interpreted broadly, and can be mechanical or electrical connection, or internal connection between two components, or direct connection. The terms up, down, left, right, etc. are only used to indicate relative positional relationships. When the absolute position of the object being described changes, the relative positional relationship may change. Secondly: The accompanying drawings of the embodiments disclosed in this invention only involve the structures involved in the embodiments disclosed in this invention. Other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of this invention can be combined with each other. In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A centralized intelligent management system for the entire lifecycle of a photovoltaic power station, characterized in that, include: A model management module is used to build and maintain a digital twin that is mapped to a physical photovoltaic power plant in real time. The model management module is configured to perform a model reconstruction operation in response to an instruction representing a life cycle phase transition. The reconstruction operation dynamically changes the set of simulated objects and the level of detail of their attributes in the digital twin. The life cycle phases include planning, construction, operation, maintenance and decommissioning phases. A data service module is used to standardize and encapsulate heterogeneous data from various life cycle stages according to a predefined data structure template corresponding to the photovoltaic power station field. The data structure template requires each data entity to be bound to a semantic type label taken from a closed type dictionary and a timestamp accurate to the second, and provides the encapsulated data stream to the model management module. A decision guidance module is used to generate strategy instructions to guide work at specific lifecycle stages based on the simulated state presented by the digital twin after inputting the encapsulated data stream. A state tracing module is used to record all external instructions, data update events and internal policy instructions received by the system in their order of occurrence as an unmodifiable linear event sequence, and to reproduce the complete state of the digital twin at any specified historical moment by replaying the event sequence sequentially from the initial state.

2. The centralized photovoltaic power plant full life cycle intelligent management system according to claim 1, characterized in that, The instructions upon which the model management module executes the model reconstruction operation are automatically triggered by any of the following conditions: the preset milestone event marker in the business process data is identified; or the decision guidance module issues a phase transition suggestion and obtains confirmation through the authorized interface; or the data service module reports a structural change in the component topology of the physical power station involving a preset percentage or more of the number of nodes.

3. The centralized photovoltaic power station full life cycle intelligent management system according to claim 2, characterized in that, The model reconstruction operation is specifically manifested as follows: S11. When the phase changes from planning to construction, in the digital twin, the set of simulated objects is expanded from the site-level geographic space and electrical boundaries to include the location coordinates and electrical connection relationships of key equipment. S12. When the phase transitions from construction to operation, the digital twin further refines the attributes of the key equipment from static parameters to include real-time operating parameters and health status indicators calculated based on performance data.

4. The centralized photovoltaic power plant full life cycle intelligent management system according to claim 1, characterized in that, When generating policy instructions, the decision guidance module calls the historical event sequence provided by the state tracing module and optimizes policy generation by executing the following steps: S21. Extract a set of key feature vectors from the current state of the digital twin; S22. Retrieve historical state points and their subsequent events with similar feature vectors in the historical event sequence; S23. Analyze the strategies executed at the historical state points and the impact of the subsequent event sequences on the long-term performance indicators of the system. S24. Based on the analysis results, adjust the generation logic of the current strategy instructions to avoid the negative effects that have been shown in the historical strategies or to enhance their positive effects.

5. The centralized photovoltaic power plant full life cycle intelligent management system according to claim 1, characterized in that, The data service module has a predefined data structure template whose closed type dictionary covers all core entity types from photovoltaic modules and inverters to construction logs and maintenance work orders. The semantic type tags ensure consistent understanding and seamless integration of data across stages at the semantic level.

6. The centralized photovoltaic power plant full life cycle intelligent management system according to claim 1, characterized in that, Before generating the final policy instruction, the decision guidance module is configured to perform the following simulated optimization steps: S31. Based on the current state of the digital twin, generate at least two candidate strategies; S32. Using each candidate strategy as an input event, drive the digital twin to perform virtual execution, and obtain a set of system status indicators after virtual execution. The indicators include at least system efficiency, security risk level and expected benefits. S33. Based on the preset priority rules, compare the system status indicators of each group to select the final output strategy instruction.

7. The centralized photovoltaic power plant full life cycle intelligent management system according to claim 6, characterized in that, The preset priority rule is configured as follows: S41. First, assess the security risk level of all candidate strategies and select a set of strategies with a risk level lower than a preset threshold. S42. In the filtered set of strategies, the candidate strategy that maximizes the improvement in system efficiency is selected as the final output.

8. The centralized photovoltaic power plant full life cycle intelligent management system according to claim 1, characterized in that, When the model management module detects that the power plant has entered the decommissioning phase, the model reconstruction operations performed include: S51, Remove the simulation of real-time operating parameters of the equipment; S52. Associate the equipment object with the asset residual value assessment model and material recycling and processing path information; S53. Shift the focus of the model's calculations from real-time power generation performance to the simulation and evaluation of the value of the inventory and the subsequent utilization plan of the site.

9. The centralized photovoltaic power station full life cycle intelligent management system according to claim 4, characterized in that, The event sequence recorded by the state tracing module constitutes the unique source of facts for the system state. Based on this, the audit query function can reverse locate and output all upstream related events that led to the result and their precise timestamps for occurrence for any policy instruction or data change, forming a complete causal chain.

Images